[petsc-users] Petsc 3.19.2 to 3.18.0 error possibility.
Barry Smith
bsmith at petsc.dev
Sat Jun 24 12:08:04 CDT 2023
Older versions of PETSc have generally less error checking and debug versions have more error checking than optimized version. Once you update the code for newer versions it should work clearly for both debug and optimized versions of PETSc.
Barry
> On Jun 24, 2023, at 12:08 PM, Paweł Stebliński <psteb at bobolin.com.pl> wrote:
>
> Welcome
>
> Thank You for Your answer.
>
> I understand the error. When You setValues in Vector, it must be assembled. The same in matrices. But exception is not thrown in 3.17.5 (no debuging switch) and is thrown from 3.18.0 to 3.19.2 (with or without debugging switch). Error is not thrown in old version of petsc 3.1-p8.
>
> Moreover the exception is thrown in 3.17.5 version ( --with-debuging=yes) but is not thrown in the version with switch --with-debuging=no. Don't You think that it doesn't matter which version is considered: with or without debbuging. The exception should be thrown in both cases.
>
> If some of the vectores were not assembled it should be thrown exception in older version 3.1-p8. But there is no exception in older version.
>
> The code adjustment which is compatible to new petsc was not done in such fundamentals details like vector acssembling. The programmer should mostly change the names of variables and functions from old to new.
>
> Best regards
>
> Paul
>
>
>
> W dniu 24.06.2023 o 16:32, Barry Smith pisze:
>>
>> Look for places where you call VecSetValues(), make sure that after the call before you use the vector for some other use you call VecAssemblyBegin/VecAssemblyEnd
>>
>>> On Jun 23, 2023, at 6:57 PM, Paweł Stebliński <psteb at bobolin.com.pl> <mailto:psteb at bobolin.com.pl> wrote:
>>>
>>> I'm using petsc 3.19.2 --with-debuging=yes (generated executable a little bigger than with no option). MPICH2 4.1.1. also --download-fblaslapack=1, --with-clanguage=C++, --with-mpi-dir=pathtompi, --with-x=0, --with-precision=DOUBLE,
>>>
>>> I have changed in PETSC_ARCH directory in file petscvariables: CLANGUAGE = from CXX to C due to fact that I am using CFLAGS variable indicating Include files in Makefile
>>>
>>> running path:
>>>
>>> nice -n -20 pathtompi/bin/mpiexec.gforker -np 80 pathtomagpar/src/magpar.exe $params
>>>
>>> OUTPUT (This output, due to distributed processing appears long after point of seqaij matrix init.):
>>>
>>> =================================================================================(START)
>>>
>>>
>>>
>>> [6]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
>>> [6]PETSC ERROR: Object is in wrong state
>>> [6]PETSC ERROR: Not for unassembled vector, did you call VecAssemblyBegin()/VecAssemblyEnd()?
>>> [6]PETSC ERROR: WARNING! There are option(s) set that were not used! Could be the program crashed before they were used or a spelling mistake, etc!
>>> [6]PETSC ERROR: Option left: name:-addcalorics value: 0 source: command line
>>> [6]PETSC ERROR: Option left: name:-addhtomeq value: 0 source: command line
>>> [6]PETSC ERROR: Option left: name:-addjtog value: 0 source: command line
>>> [6]PETSC ERROR: Option left: name:-addterm value: 0 source: command line
>>> [6]PETSC ERROR: Option left: name:-condinp_j value: 1e99 source: file
>>> [6]PETSC ERROR: Option left: name:-condinp_t value: 1e-4 source: command line
>>> [6]PETSC ERROR: Option left: name:-countG value: 0 source: command line
>>> [6]PETSC ERROR: Option left: name:-countN value: 0 source: file
>>> [6]PETSC ERROR: Option lef[8]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
>>>
>>>
>>>
>>> ================================================================================(END)
>>>
>>>
>>>
>>>
>>>
>>> when you set: --with-debuging=no and add to code:
>>>
>>> ierr = PetscMallocSetDebug(PETSC_TRUE,PETSC_TRUE);CHKERRQ(ierr); in main.c file before matrix initialization.
>>>
>>> You get OUTPUT BELOW: (about the point where function initialize matrix).
>>>
>>> When you comment matrix initialization the exception doesn't appear.
>>>
>>>
>>> [0]PETSC ERROR: PetscTrFreeDefault() called from PetscEventRegLogRegister() at /home/psteb/PRACA/LIB_NEW/petsc-3.19.2/src/sys/logging/utils/eventlog.c:363
>>> [0]PETSC ERROR: Block at address 0x55a6d104bbb0 is corrupted; cannot free;
>>> may be block not allocated with PetscMalloc()
>>> [0]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
>>> [0]PETSC ERROR: Memory corruption: https://petsc.org/release/faq/#valgrind
>>> [0]PETSC ERROR: Bad location or corrupted memory
>>> [0]PETSC ERROR: WARNING! There are option(s) set that were not used! Could be the program crashed before they were used or a spelling mistake, etc!
>>> [0]PETSC ERROR: Option left: name:-addcalorics value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-addhtomeq value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-addjtog value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-addterm value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-condinp_j value: 1e99 source: file
>>> [0]PETSC ERROR: Option left: name:-condinp_t value: 1e-4 source: command line
>>> [0]PETSC ERROR: Option left: name:-countG value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-countN value: 0 source: file
>>> [0]PETSC ERROR: Option left: name:-D value: 20.0 source: file
>>> [0]PETSC ERROR: Option left: name:-demag value: 0 source: command line
>>> [0]PETSC ERROR: Option left: name:-dmi value: 2 source: command line
>>> [0]PETSC ERROR: Option left: name:-dmi_xyzfile value: dmi.xyz source[server]: PMIU_parse_keyvals: unexpected key delimiter at character 48 in cmd
>>>
>>>
>>> IN POLAND is midnight so I go sleep. Tommorow i will check --with-cflags='-g -O0' --with-cxxflags='-g -O0' --with-fflags='-g -O0'
>>>
>>> Best Regards
>>>
>>> Paul Steblinski
>>>
>>>
>>>
>>>
>>>
>>> W dniu 23.06.2023 o 23:41, Barry Smith pisze:
>>>>
>>>> Could you send us the exact error output that occurs? Cut and paste the run command and the entire error message.
>>>>
>>>> Also send the configure options you used. Have you tried configuring the later PETSc versions with all optimization turned off; use --with-debugging=1 --with-cflags='-g -O0' --with-cxxflags='-g -O0' --with-fflags='-g -O0' Does the same error occur?
>>>>
>>>> Barry
>>>>
>>>>
>>>>
>>>>
>>>>> On Jun 23, 2023, at 5:30 PM, Paweł Stebliński via petsc-users <petsc-users at mcs.anl.gov> <mailto:petsc-users at mcs.anl.gov> wrote:
>>>>>
>>>>> I am micromagnetic (MAGPAR) software developer. Old Magpar version has been using petsc 3.1-p8. I have decided to upgrade to petsc 3.19.2 with avx512 support. Unfortunately there appeared an error during software testing. Error appeares in ranning code after proper compiling and linking. The bug is in a code part which initializes matrix in petsc library versions: 3.19.2, 3.19.1, 3.19.0 and also from 3.18.5 to 3.18.0. If we use petsc version 3.17.5 the error doesn't appear. With this version (3.17.5) all is ok and simulation is running without any errors or throwing exceptions. My guess is linked to avx512 implementation which is good up to 3.17.5 version and buggy in upper mentioned versions with higher numbers. Avx512 is buggy according to tested SeqAij matrices.
>>>>>
>>>>> The exception is not thrown if we comment code fragment below.
>>>>>
>>>>> ierr = MatCreateSeqAIJ(
>>>>> PETSC_COMM_SELF,
>>>>> nvert,nvert,
>>>>> 0,ia,
>>>>> &mat
>>>>> );CHKERRQ(ierr);
>>>>> ierr = MatSetFromOptions(mat);CHKERRQ(ierr);
>>>>>
>>>>> ia - is number of nonzeros array which is obtained according parmetis partitioning. There were the same version of parmetis (3.1.1) in the all considered cases.
>>>>>
>>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20230624/ed443f18/attachment-0003.html>
More information about the petsc-users
mailing list